The private repo as the real agent workspace
A sharp example of open source as an export surface, while the valuable development asset becomes the private agent-readable wiki, specs, tests, and experiment harness.
Links: Original source · Related link 1 · Related link 2 · Related link 3
Logged at IST: 2026-08-25 12:35 IST
What it is: Ryan Dahl describing how celld is developed from a larger private denoland/celld.dev repo, while the public denoland/celld repo stays a lean export.
Gist: The useful part is the shape of the working repo. Dahl says the private repo has an LLM-maintained design wiki, a large test suite, a TLA+ spec, website code, cloud experiment scripts, and a speculative managed control plane. The public repo is open source, but the real development loop happens around the richer private artifact.
That feels like an agent-era pattern: code is no longer the only scarce asset. The durable value is the executable and readable context around the code: specs, tests, design pages, operational scripts, and experiment harnesses. If those are strong enough, agents can keep rewriting, porting, or replacing large chunks of the app while the test/spec corpus acts as the contract.
It also changes the open-source boundary. Instead of sharing every commit, a project can publish a clean artifact and keep the private repo as the agent workspace where product strategy, control-plane ideas, tests, and design history live together.
Newsletter angle: Strong developer-tools item: in agent-heavy development, the private test/wiki/spec corpus may become the real moat, and the public repository becomes more like a release artifact.